[FEATURE REQ] Support `reasoning` in ChatCompletion API (CompletionStreamResponseDelta)

[gaspardpetit]

Describe the feature or improvement you are requesting

Hello - I understand that reasoning content are not officially part of the chat completion API, but they are supported by the chat completion provided by ollama, and work specifically with OpenAI's gpt-oss models.

https://github.com/ollama/ollama/blob/main/openai/openai.go

type Message struct {
	Role       string     `json:"role"`
	Content    any        `json:"content"`
	Reasoning  string     `json:"reasoning,omitempty"`
	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`
	Name       string     `json:"name,omitempty"`
	ToolCallID string     `json:"tool_call_id,omitempty"`
}

The chat completion api of ollama will emit elements with reasoning deltas, in addition to the ones with content.

This can be supported in the C# client with minimal changes to InternalChatCompletionStreamResponseDelta.cs and InternalChatCompletionStreamResponseDelta.Serialization.cs

Thank you for considering adding support for this.

Additional context

No response

[jsquire]

Hi @gaspardpetit. Thank you for reaching out and for your suggestion. The OpenAI client library is intentionally modeled for the official OpenAI REST API. While we do not discourage use with other services/models, neither do we offer support for them. As you noted, the reasoning field is not part of the documented chat streaming response nor does it appear in the OpenAI REST API specification. As a result, this is not a field that we'll be adding to the client.

There are two options available for you to consume the additional data. Starting in v2.6.0, you're able to use the JsonPatch feature to read/write additional data on models in the chat update:

var client = new ChatClient("gpt-5", "<< YOUR API KEY >>");
var stream = client.CompleteChatStreamingAsync(ChatMessage.CreateUserMessage("You are Terry Crews and have had 7 cans of RedBull this morning.  Tell me the story of the 3 little pigs."));

await foreach (var update in stream)
{
    var reasoning = update.Patch.GetString("$.reasoning"u8);
}

You're also able to drop down to the protocol level and work with the raw SSE data directly if you needed greater control. An example using the high-level shape of the Go struct (while skipping some details around content and tool calls) would look something like:

var client = new ChatClient("gpt-5", "<< YOUR API KEY >>");
var stream = client.CompleteChatStreamingAsync(ChatMessage.CreateUserMessage("You are Terry Crews and have had 7 cans of RedBull this morning.  Tell me the story of the 3 little pigs."));

await foreach (var page in stream.GetRawPagesAsync())
{
    var parser = SseParser.Create(page.GetRawResponse().ContentStream!);
    
    foreach (var item in parser.Enumerate())
    {
        var message = JsonSerializer.Deserialize<Message>(item.Data);
        var reasoning = message.Reasoning;
    }
}

public record Message(
    string Role,
    object Content,
    string Reasoning,
    object ToolCalls,
    string Name,
    string ToolCallId);